Abstract
Accelerometers with low sampling rates (1 Hz) are commercially available as ear tags. While an automated and therefore undisturbed sampling of animal behaviour can be useful not only in behavioural studies but also in ecological or wildlife management studies, the usefulness of such 'a low data collection rate for the prediction of behaviours was the key question addressed here. We classified the behaviour of female wild boar, kept under semi-natural conditions in a large outdoor enclosure, using acceleration data. Predictions were based on a machine learning algorithm, specifically a random forest model in the open software h2o. Remarkably, prediction of many behaviours was possible using ear-tag acceleration sensors that sampled data only at a low frequency. This measurement device was mainly used to minimise the potentially harmful effects caused by the repeated capture of wild animals to exchange batteries. Long battery life will also help to collect long-term accelerometer data and has the potential to explore seasonal and inter-annual trends. Foraging, lateral resting, sternal resting and lactating were identified well, scrubbing, standing and walking not reliably. Balanced accuracy depended on the behaviour type and ranged from 50% (walking) to 97% (lateral resting). Results show that static features of unfiltered acceleration data, as well as of gravitation and orientation filtered data, were used in the prediction of behaviour. The waveform of certain behaviours in the sampled frequency range played no important role. Certain positively identified behaviours, such as food intake and lactation, could be of interest for wildlife managers attempting to control population growth in this pest-species. We provide several R-scripts that allow the analysis of behavioural accelerometer data.